Input: $\mathcal{O},\omega ,\mu$

1:  Randomly initialize A, B, C.

2:  repeat

3:         ${\mathbf{A}}_{i:}=\max (\mathrm{0,}{\mathbf{A}}_{i:}-{\nabla }^{2}f{({\mathbf{A}}_{i:})}^{-1}\nabla f({\mathbf{A}}_{i:}))$ for all i.

4:         Update θ for logistic regression

5:         $vec(\mathbf{B})=\max (\mathrm{0,}vec(\mathbf{B})-{\nabla }^{2}g{(vec(\mathbf{B}))}^{-1}\nabla g(vec(\mathbf{B})))$.

6:         $vec(\mathbf{C})=\max (\mathrm{0,}vec(\mathbf{C})-{\nabla }^{2}g{(vec(\mathbf{C}))}^{-1}\nabla g(vec(\mathbf{C})))$.

7:   until Converged

8:   ${\overline{\mathbf{A}}}_{:r}\leftarrow \frac{{\mathbf{A}}_{:r}}{\left|\right|{\mathbf{A}}_{:r}\left|\right|}$, ${\overline{\mathbf{B}}}_{:r}\leftarrow \frac{{\mathbf{B}}_{:r}}{\left|\right|{\mathbf{B}}_{:r}\left|\right|}$, ${\overline{\mathbf{C}}}_{:r}\leftarrow \frac{{\mathbf{C}}_{:r}}{\left|\right|{\mathbf{C}}_{:r}\left|\right|}$, $\forall r$

9:   ${\overline{\mathbf{A}}}_{ir}\leftarrow 0\mathrm{if}{\overline{\mathbf{A}}}_{ir}<{10}^{-6}$, ${\overline{\mathbf{B}}}_{jr}\leftarrow 0\mathrm{if}{\overline{\mathbf{B}}}_{jr}<{10}^{-3}$, ${\overline{\mathbf{C}}}_{kr}\leftarrow 0\mathrm{if}{\overline{\mathbf{C}}}_{kr}<{10}^{-3}\forall i,j,k,r$

10:   return  $\mathcal{X}={\sum }_{r=1}^R{\lambda }_r{\overline{\mathbf{A}}}_{:r}{\overline{\mathbf{B}}}_{:r}{\overline{\mathbf{C}}}_{:r}$.